Machine Learning-Enabled Prediction of Synthesis–Property Relationships in Cobalt Nanowires as Rare-Earth-Free Permanent Magnets

The development of rare-earth-free permanent magnetic materials has become increasingly important due to concerns about supply chain vulnerabilities and sustainability associated with rare-earth elements. Cobalt nanowires represent promising candidates for high-performance permanent magnets, yet understanding the complex relationships between the synthesis parameters and magnetic properties remains challenging. This study establishes quantitative synthesis–property relationships for cobalt nanowires through machine learning analysis and experimental validation. A comprehensive database containing complete synthesis–property records was constructed from the peer-reviewed literature focusing on polyol reduction methods. Among the 18 evaluated machine learning algorithms, gradient-boosting decision trees demonstrated superior predictive performance for coercivity prediction. Shapley Additive exPlanations analysis identified reducing agents as the most critical parameter influencing coercivity, followed by nucleating agents and process parameters. To validate these machine learning insights, systematic experimental investigations were conducted using four different reducing agents, while maintaining identical synthesis conditions. First-order reversal curve analysis further validated the critical role of reducing agents in determining the magnetic uniformity and switching behavior. This work demonstrates the effectiveness of machine learning in elucidating synthesis–property relationships and provides experimental validation of model predictions, establishing a foundation for the rational design of high-performance cobalt nanowires.